Other Eigencuts points

[Shannon Quinn <sq...@gatech.edu>]

Hi all,

These are a couple side-items that still need to be resolved with Eigencuts
that don't really cleanly fit into the existing JIRA tickets (516, 517, and
518).

-Related to the issue of 518, if we want Eigencuts to support non-image raw
input, that begs the ability to process non-symmetric data (i.e. data for
which the underlying graph is not necessarily bi-directional between every
pair of nodes). There are ways of "converting" non-symmetric to symmetric
via some approximations (some of which are mentioned in the original
Eigencuts paper, and others that folks on this list are probably very
knowledgeable in), so that might be a task I could implement.

-There's an interesting conflict of data types in the current implementation
that I haven't been able to come up with a fix for yet: a spot where I have
to convert a List<Double> object to a Mahout Vector. The List contains the
eigenvalues from the Lanczos eigensolver, and the Vector is needed for
serializing to HFS during some later M/R tasks that require the eigenvalues.
In looking at the eigensolver code where the List is originally created, it
isn't obvious to me that the list could even be safely built as a Vector
instead, much less how any other code might be dependent on that data. Is
there a way to solve this problem without my current "hack" of simply
pasting each element in the list one-by-one into a Vector?

-If anyone is going to ApacheCon who's never been to Atlanta before, I'd be
happy to act as a guide (my old stomping grounds and such) :)

Shannon

Re: Other Eigencuts points

[Ted Dunning <te...@gmail.com>]

This is, as they say, picking fly-specks out of the pepper.

THere is no way you could even measure the difference.

On Tue, Oct 5, 2010 at 9:57 AM, Jeff Eastman <jd...@windwardsolutions.com>wrote:

>  Even if the list is a couple hundred, the conversion time will be dwarfed
> by the iterations. I'm +1 to disregard unless/until we can prove this to be
> a bottleneck.
>
>
> On 10/5/10 5:34 AM, Shannon Quinn wrote:
>
>> I only wonder because the algorithm is iterative, so even if the list of
>> eigenvalues is short, the conversion will still happen every time the
>> algorithm loops. But if this is small enough to disregard, that's totally
>> fine.
>>
>> Apologies for the brevity, this was sent from my iPhone
>>
>> On Oct 5, 2010, at 3:22, Ted Dunning<te...@gmail.com>  wrote:
>>
>>  That isn't such a big deal.  Just a short loop.
>>>
>>> On Mon, Oct 4, 2010 at 7:15 PM, Shannon Quinn<sq...@gatech.edu>  wrote:
>>>
>>>  Is
>>>> there a way to solve this problem without my current "hack" of simply
>>>> pasting each element in the list one-by-one into a Vector?
>>>>
>>>>
>

Re: Other Eigencuts points

[Jeff Eastman <jd...@windwardsolutions.com>]

  Even if the list is a couple hundred, the conversion time will be 
dwarfed by the iterations. I'm +1 to disregard unless/until we can prove 
this to be a bottleneck.

On 10/5/10 5:34 AM, Shannon Quinn wrote:
> I only wonder because the algorithm is iterative, so even if the list of eigenvalues is short, the conversion will still happen every time the algorithm loops. But if this is small enough to disregard, that's totally fine.
>
> Apologies for the brevity, this was sent from my iPhone
>
> On Oct 5, 2010, at 3:22, Ted Dunning<te...@gmail.com>  wrote:
>
>> That isn't such a big deal.  Just a short loop.
>>
>> On Mon, Oct 4, 2010 at 7:15 PM, Shannon Quinn<sq...@gatech.edu>  wrote:
>>
>>> Is
>>> there a way to solve this problem without my current "hack" of simply
>>> pasting each element in the list one-by-one into a Vector?
>>>

Re: Other Eigencuts points

[Shannon Quinn <sq...@gatech.edu>]

I only wonder because the algorithm is iterative, so even if the list of eigenvalues is short, the conversion will still happen every time the algorithm loops. But if this is small enough to disregard, that's totally fine. 

Apologies for the brevity, this was sent from my iPhone

On Oct 5, 2010, at 3:22, Ted Dunning <te...@gmail.com> wrote:

> That isn't such a big deal.  Just a short loop.
> 
> On Mon, Oct 4, 2010 at 7:15 PM, Shannon Quinn <sq...@gatech.edu> wrote:
> 
>> Is
>> there a way to solve this problem without my current "hack" of simply
>> pasting each element in the list one-by-one into a Vector?
>> 

Re: Other Eigencuts points

[Ted Dunning <te...@gmail.com>]

That isn't such a big deal.  Just a short loop.

On Mon, Oct 4, 2010 at 7:15 PM, Shannon Quinn <sq...@gatech.edu> wrote:

> Is
> there a way to solve this problem without my current "hack" of simply
> pasting each element in the list one-by-one into a Vector?
>